VRIFY.AI represents a paradigm shift in mineral exploration, leveraging the power of advanced algorithmic models to analyze and interpret complex exploration datasets. By integrating vision transformer models with sophisticated machine learning classifiers, we generate probabilistic 3D maps of the Earth's subsurface, aimed at pinpointing prospective mineral deposits.

Our system is trained on extensive datasets, including both public geological information and proprietary client data, ensuring a comprehensive understanding of each exploration context. The confidentiality of your data is paramount, and our models are customized to your specific project requirements without compromising data privacy.

What sets VRIFY.AI apart is its ability to augment your existing geological expertise through informed, objective and unbiased analysis of multiple data dimensions, significantly enhancing the prospectivity results and thereby reducing the risk and cost associated with exploration drilling.

In essence, VRIFY.AI offers a data-driven, highly secure, and efficient approach to mineral exploration, enabling your team to make informed decisions and prioritize exploration targets with a higher degree of confidence and lower environmental impact.

Introduction To Drill Targeting

Data Management

Privacy Considerations

Understanding the Outputs

VRIFY.AI operates by merging the capabilities of image transformer architectures (vit) with machine learning classification techniques to produce probabilistic representations of the Earth in three-dimensional space. A prediction for the VRIFY Prospectivity Score (VPS) is made by training a supervised learning model from the existing mineral occurrences and embedded data space created by the transformers. For each area of interest, data is compiled, and a uniquely fine-tuned and optimized model is developed, tailored to the specifics of the available information.

Our methodology in developing these models is grounded in the principle of unbiased objectivity, ensuring that the machine learning approaches we employ are as neutral as possible. We deliberately avoid incorporating subjective data or relying on human feedback loops for model enhancement. Instead, our strategy is to allow the data to guide the learning process autonomously. This approach helps in mitigating potential cognitive and confirmation biases that might arise from preconceived notions or human influence, ensuring that the models develop insights based purely on the inherent patterns and relationships present within the data itself. By prioritizing data-driven learning, we aim to create models that offer reliable, objective, and accurate predictions, reflecting a true understanding of the underlying geologic phenomena without the skew of subjective interpretation.

While human geologists excel at recognizing patterns and leveraging their expertise for predictions, their ability to process complex patterns is typically limited to 3 to 4 layers of information simultaneously. Beyond this threshold, the effectiveness in identifying patterns diminishes greatly. Machines, by means of statistical models, however, can analyze data across multiple dimensions and utilize their flawless memory to reference previously encountered information.

Artificial Intelligence (AI) learning models significantly enhance our capacity to scrutinize exploration data, offering more objective, unbiased and result-driven predictions. AI not only supports the validation of existing targets but also identifies potential ones that might have been overlooked. This heightened accuracy in target identification reduces the likelihood of drilling unnecessary holes, thereby improving the overall chances of successful discovery. Furthermore, by incorporating both positive and negative results from exploration work, the model can benefit from both successful holes and misses.

Machine learning is a subfield of Artificial intelligence, but both rely on mathematical processes. All current algorithms in the public domain are underlain by not so advanced mathematical operations, the only reason for their growth in recent years is the drastic increase in computing power. Even the most advanced architecture could be calculated by hand if you had enough time!

Now is the time to consider AI targeting across all stages of mining, from initial exploration through resource delineation to the management of operating mines. Leveraging AI for informed decision-making can save millions in exploration costs and significantly reduce wasted time at every phase. Swift project development and rapid resource delineation, underpinned by AI at each step, benefit financial outcomes, environmental sustainability, and local communities alike.

We saw an opportunity to revolutionize client outcomes and fundamentally transform how we meet our clients' needs. Our motivation stems from a passion for breaking new ground, embracing technological advancements, and dreaming big. We recognize the critical importance of minerals to our society and understand the challenges associated with accelerating project developments and facilitating more efficient discoveries. By entering the AI mineral system targeting arena, VRIFY aims to catalyze investment in this space, demonstrating our commitment to innovative solutions that address the mining industry's evolving demands.

In order to generate predictions, VRIFY needs learning examples (mineral occurrences, drill hole assays, rock geochemistry, etc.) and some exploration data.

All sources of exploration data can potentially be utilized in training the model and making predictions. The models can handle geophysical datasets (Gravity, IP, Magnetics, EM, MT), terrain models (Digital Elevation Models), mapping points, structural measurements and analysis, geochemical datasets together with downhole data.

For AI and machine learning models, having a larger dataset generally leads to better outcomes. However, challenges can arise from datasets that are sparse or lack overlap, potentially affecting the model's accuracy. VRIFY evaluates the available data volume prior to entering any agreement, ensuring that we can deliver actionable, data-driven insights that truly benefit our clients. Furthermore, since our predictive models are trained on public and private datasets, we can mitigate the impact of your sparse datasets.

We enhance the accuracy of our predictions by integrating publicly available data into our models. This includes information from geological surveys, mineral and mining databases, along with data released by both public and private entities.

The combination of novel interpolation and data encoding approaches enables VRIFY to “fill in” data gaps in the exploration data. This is necessary since exploration datasets can be on different supports, some data are associated to grids and other to 3D spatial points. In order to combine all the information on a similar support, novel approaches needed to be developed for data interpolation and predictive modeling.

Upon receiving the exploration data, the VRIFY team will conduct a careful review, QAQC and compilation process. Once these steps are completed, the data will be validated with the technical team from the participating company. This step is crucial in making sure the data is of the highest possible quality. Most often, this initial compilation is the longest and represents 80% of the amount of work to be conducted. Depending on the level of organization of the transferred data this can take from days to weeks. Once the data is compiled and transferred to the proper formats, the training and predictions usually take a few days. Once the model is trained, updates can be made within a few hours of receiving new data.

One of the main drivers for the development of this targeting system was to build a living/breathing model. Oftentimes, exploration and targeting models are static and rarely updated more than once or twice a year. Our approach is to construct a predictive model that can be updated quickly and provide instant feedback on your ongoing exploration activities.

Once a pre-trained model is available for your deposit type and area of interest (AOI), updates can be made by integrating new exploration results to the learning features. These results can be from the company's exploration efforts, such as drilling or prospecting. We can also incorporate data from neighboring projects in order to refine the local predictions for your AOI.

Updates can be made as often as required to help you move your exploration project forward.

VRIFY consistently upholds the highest standards of client confidentiality. Similar to our visualization data, the AI-transferred data is subject to stringent policies, procedures, and SOC2 compliance throughout the entire data pipeline, ensuring the utmost security and privacy.

The final product is a grid of probabilities. This is the VPS scoring that can be displayed within our viewing platform. This will also be delivered as a 3D point cloud that can be imported in by third party geoscientific software.

The accuracy of the model hinges on both the quantity and the quality of the input data. Our data team excels in cleaning and processing data to ensure the highest quality inputs for modeling. The principle "garbage in, garbage out" remains as relevant as ever, emphasizing the importance of reliable data for achieving accurate model outcomes. As part of the training process, a portion of the learning data is set aside as testing data. The model is trained without these points and then applied to make a prediction on those points. Since the outcome of those points is known, it is possible to evaluate the accuracy of the predictive model.

Our models are versatile and applicable to all deposit types, yet they are specially tailored for distinct groups of commodities. The training process and input features are derived from careful mineral system analysis in order to construct a solid foundation to the predictions. For instance, the model used to identify lithium-bearing pegmatites differs from the one employed for locating shear-hosted gold deposits, ensuring optimal effectiveness and accuracy for each specific exploration target.

3D predictions are made by incorporating the coordinates (X,Y,Z) as part of the training dataset. The algorithm is trained to not only predict the probability of mineralization (VPS) but also the elevation of the occurrence. By combining both the VPS and the elevation prediction VRIFY is able to generate accurate 3D representations of the targets.

Since predictions are made as a probability, there is never a 100% chance of finding mineralization. However, even in the case of unsuccessful drill holes, the data can be fed into the algorithm to refine the predictive modeling. Since the algorithm learns from both positive and negative exploration examples, all the data has value.

By leveraging your entire data stack and getting unbiased predictions for our platform, you’ll ensure that you are extracting the most out of the exploration data.

Reach out to your dedicated VRIFY AI Contact or email Support@VRIFY.com for more information.